Yoav Goldberg: Every popular benchmark will be brute-forced away — so how do we measure real progress?

yoavgo · x · 2026-09-04

NLP researcher Yoav Goldberg offers a sober take on measuring AI progress: any task you can reliably score will be solved if you throw enough money ("an obscene amount") at brute-forcing its training; every challenging and popular benchmark will meet that bar. The catch — abilities won't necessarily transfer beyond the benchmark. His closing question: how can we actually measure progress?

Related event: Researchers argue benchmarks fall to brute force; real progress should be measured by "discoveries"(6 posts)→

Original post →

More from AGI Musings

• 3 months testing whether Anthropic's internal research model can design scorpion antivenom — apoorvasriniva · 2026-10-21
• Nat Lambert: AI cyber risk is an ecosystem issue, not an open-weights problem — natolambert · 2026-10-21
• Obesity drugs still out-earn OpenAI and Anthropic combined at $71B a year — HealthcareAIGuy · 2026-10-21
• AI can simulate how a self talks, but not what a self is — gerardsans · 2026-10-21
• Musk predicts money will become irrelevant, replaced by 'wattage and tonnage' — Polymarket · 2026-10-21
• zetalyrae quips: post-singularity utopia is so great you need brain surgery to cope — wfithian · 2026-10-21

AGI Musings channel →